Bakas, Spyridon

Abstract INTRODUCTION Glioblastoma is the most common malignant adult brain tumor with poor prognosis, largely due to its heterogeneous landscape. The AI-RANO and RANO-RGP groups organized the BraTS-Pathology 2024 challenge, providing a publicly-available dataset and a benchmarking environment for AI models to automatically identify six clinically-relevant histopathologic sub-regions in H&E-stained whole slide images (WSIs). METHODS We reclassified the TCGA-GBM and TCGA-LGG data according to the WHO 2021 criteria and identified diagnostic WSIs of 188 glioblastoma (IDH-wt, CNS WHO Gr.4) patients. A well-defined annotation protocol was communicated to eight neuropathologists, across eight institutions, who created “ground truth” annotations for Cellular Tumor (CT), Geographic Necrosis, Cortical Infiltration, Pseudopalisading Necrosis (PN), Microvascular Proliferation, and Penetration into White Matter (PWM). Comprehensive WSI curation for elimination of artifactual content and follow up partitioning resulted in approximately 300,000 data samples/patches. FINDINGS Inter-observer variability assessment highlighted difficulties in establishing a “gold standard”. The least annotated sub-regions (PWM and PN) were the most challenging to the participating methods, whereas all methods achieved their highest accuracy in the most annotated sub-region (CT). The highest average performance across sub-regions was obtained by a foundation model-based method. However, this top-ranked solution achieved an F1 score of 0.51 (Accuracy=0.65, Sensitivity=0.56, Specificity=0.93). CONCLUSIONS Our findings suggest that achieving state-of-the-art performance, in automatically identifying clinically-relevant glioblastoma sub-regions, remains an unsolved research problem. Analysis of the ground-truth annotations underscores the critical need for more precisely defined and neuropathologically standardized annotation protocol. Current AI algorithms are facing difficulties in addressing glioblastoma’s heterogeneous histologic landscape in the absence of large, comprehensively, and consistently annotated data. Building upon these findings, and towards developing more accurate AI algorithms, the 2025 iteration of the BraTS-Path challenge is currently on-going by leveraging two million data samples, aiming to enhance our disease understanding and improve diagnostic accuracy, ultimately contributing to improved patient outcomes.

Abstract BACKGROUND Volumetric analysis of brain metastatic lesions is of great relevance in neuro-oncology. Since manual annotation of a lesion is a laborious and time-consuming task, utilizing artificial intelligence (AI) tools for automated segmentation of brain metastases has shown promising potential, but published algorithms are not incorporated into clinical practice because of poor generalizability across institutions. Large publicly available and well-annotated datasets from across the world are critical for further clinical translation. MATERIAL AND METHODS The Brain Tumor Segmentation - Metastases (BraTS-METS) 2025 Lighthouse Challenge provides a competitive environment where scientists from various disciplines can benchmark their AI-based algorithms for automatic segmentation of pre- and post-treatment brain metastases on a well-curated publicly available brain metastasis dataset comprising 1,778 multiparametric brain MRI cases. Each case consists of T1-weighed, T1 post-contrast and FLAIR sequences. The cases were retrospectively obtained from 8 institutions in the USA and overseas, and underwent 3D nnU-Net pre-segmentation, followed by manual refinements performed by medical students and neuroradiologists, and a final approval by two independent board-certified neuroradiologists. A testing dataset was prepared through four rounds of annotation: two completed ‘from scratch’ and two following 3D nnU-Net pre-segmentation, all conducted by two independent neuroradiologists, with the segmentation process screen recorded for future analysis. This allowed us to establish inter- and intra-rater variability in brain metastases segmentation and create a reference standard dataset that will be publicly available after the conclusion of the challenge. The model performance of the participating teams in the challenge will be assessed via the following metrics: Dice Similarity Coefficient (DSC), Normalized Surface Distance (NSD), sensitivity, and precision of the model. RESULTS A multi-annotator testing dataset of 75 MRI cases of brain metastases and 300 segmentations (150 created ‘from scratch’ and 150 from 3D nnU-Net pre-segmentations of the cases) was developed as a new reference standard. The mean intra- and inter-rater DSC for the segmentations ‘from scratch’ were 0.68 and 0.59, respectively. The mean intra- and inter-rater DSC for the segmentations built upon 3D nnU-Net pre-segmentations were 0.88 and 0.77, respectively. CONCLUSION The BraTS-METS 2025 Lighthouse Challenge is a key initiative designed to advance the use of AI to improve characterization of brain metastases by providing algorithms for automatic segmentation of all lesions and peritumoral edema that are measured against a novel reference standard based on rigorous multi-annotator brain MRI segmentations.

Pediatric central nervous system tumors are the leading cause of cancer-related deaths in children. The five-year survival rate for high-grade glioma in children is less than 20%. The development of new treatments is dependent upon multi-institutional collaborative clinical trials requiring reproducible and accurate centralized response assessment. We present the results of the BraTS-PEDs 2023 challenge, the first Brain Tumor Segmentation (BraTS) challenge focused on pediatric brain tumors. This challenge utilized data acquired from multiple international consortia dedicated to pediatric neuro-oncology and clinical trials. BraTS-PEDs 2023 aimed to evaluate volumetric segmentation algorithms for pediatric brain gliomas from magnetic resonance imaging using standardized quantitative performance evaluation metrics employed across the BraTS 2023 challenges. The top-performing AI approaches for pediatric tumor analysis included ensembles of nnU-Net and Swin UNETR, Auto3DSeg, or nnU-Net with a self-supervised framework. The BraTS-PEDs 2023 challenge fostered collaboration between clinicians (neuro-oncologists, neuroradiologists) and AI/imaging scientists, promoting faster data sharing and the development of automated volumetric analysis techniques. These advancements could significantly benefit clinical trials and improve the care of children with brain tumors.

The translation of AI-generated brain metastases (BM) segmentation into clinical practice relies heavily on diverse, high-quality annotated medical imaging datasets. The BraTS-METS 2023 challenge has gained momentum for testing and benchmarking algorithms using rigorously annotated internationally compiled real-world datasets. This study presents the results of the segmentation challenge and characterizes the challenging cases that impacted the performance of the winning algorithms. Untreated brain metastases on standard anatomic MRI sequences (T1, T2, FLAIR, T1PG) from eight contributed international datasets were annotated in stepwise method: published UNET algorithms, student, neuroradiologist, final approver neuroradiologist. Segmentations were ranked based on lesion-wise Dice and Hausdorff distance (HD95) scores. False positives (FP) and false negatives (FN) were rigorously penalized, receiving a score of 0 for Dice and a fixed penalty of 374 for HD95. The mean scores for the teams were calculated. Eight datasets comprising 1303 studies were annotated, with 402 studies (3076 lesions) released on Synapse as publicly available datasets to challenge competitors. Additionally, 31 studies (139 lesions) were held out for validation, and 59 studies (218 lesions) were used for testing. Segmentation accuracy was measured as rank across subjects, with the winning team achieving a LesionWise mean score of 7.9. The Dice score for the winning team was 0.65 ± 0.25. Common errors among the leading teams included false negatives for small lesions and misregistration of masks in space. The Dice scores and lesion detection rates of all algorithms diminished with decreasing tumor size, particularly for tumors smaller than 100 mm3. In conclusion, algorithms for BM segmentation require further refinement to balance high sensitivity in lesion detection with the minimization of false positives and negatives. The BraTS-METS 2023 challenge successfully curated well- annotated, diverse datasets and identified common errors, facilitating the translation of BM segmentation across varied environments and providing the tools for future development of personalized volumetric reports to patients undergoing BM treatment.

We describe the design and results from the BraTS 2023 Intracranial Meningioma Segmentation Challenge. The BraTS Meningioma Challenge differed from prior BraTS Glioma challenges in that it focused on meningiomas, which are typically benign extra-axial tumors with diverse radiologic and anatomical presentation and a propensity for multiplicity. Nine participating teams each developed deep-learning automated segmentation models using image data from the largest multi-institutional systematically expert annotated multilabel multi-sequence meningioma MRI dataset to date, which included 1000 training set cases, 141 validation set cases, and 283 hidden test set cases. Each case included T2, FLAIR, T1, and T1Gd brain MRI sequences with associated tumor compartment labels delineating enhancing tumor, non-enhancing tumor, and surrounding non-enhancing FLAIR hyperintensity. Participant automated segmentation models were evaluated and ranked based on a scoring system evaluating lesion-wise metrics including dice similarity coefficient (DSC) and 95% Hausdorff Distance. The top ranked team had a lesion-wise median dice similarity coefficient (DSC) of 0.976, 0.976, and 0.964 for enhancing tumor, tumor core, and whole tumor, respectively and a corresponding average DSC of 0.899, 0.904, and 0.871, respectively. These results serve as state-of-the-art benchmarks for future pre-operative meningioma automated segmentation algorithms. Additionally, we found that 1286 of 1424 cases (90.3%) had at least 1 compartment voxel abutting the edge of the skull-stripped image edge, which requires further investigation into optimal pre-processing face anonymization steps.

Context.— Distinguishing between renal oncocytic tumors, such as renal oncocytoma (RO), and a subset of tumors with overlapping characteristics, including the recently identified low-grade oncocytic tumor (LOT), can present a diagnostic challenge for pathologists owing to shared histopathologic features. Objective.— To develop an automatic computational classifier for stratifying whole slide images of biopsy and resection specimens into 2 distinct groups: RO and LOT. Design.— A total of 269 whole slide images from 125 cases across 6 institutions were collected. A weakly supervised attention-based multiple-instance–learning deep learning (DL) model was trained and initially evaluated through 5-fold cross validation with case-level stratification, followed by validation using an independent holdout data set. Quantitative performance evaluation was based on accuracy and the area under the receiver operating characteristic curve (AUC). Results.— The developed model data set yielded generalizable performance, with a 5-fold average testing accuracy of 84% (AUC = 0.78), and a closely aligning accuracy of 83% (AUC = 0.92) on the independent holdout data set. Conclusions.— The proposed artificial intelligence approach contributes toward a comprehensive solution for addressing commonly encountered renal oncocytic neoplasms, encompassing well-established entities like RO along with the challenging “gray zone” LOT, thereby proving applicable in clinical practice.

PURPOSE Breast density is a widely established independent breast cancer risk factor. With the increasing utilization of digital breast tomosynthesis (DBT) in breast cancer screening, there is an opportunity to estimate volumetric breast density (VBD) routinely. However, current available methods extrapolate VBD from two-dimensional (2D) images acquired using DBT and/or depend on the existence of raw DBT data, which is rarely archived by clinical centers because of storage constraints. METHODS We retrospectively analyzed 1,080 nonactionable three-dimensional (3D) reconstructed DBT screening examinations acquired between 2011 and 2016. Reference tissue segmentations were generated using previously validated software that uses 3D reconstructed slices and raw 2D DBT data. We developed a deep learning (DL) model that segments dense and fatty breast tissue from background. We then applied this model to estimate %VBD and absolute dense volume (ADV) in cm 3 in a separate case-control sample (180 cases and 654 controls). We created two conditional logistic regression models, relating each model-derived density measurement to likelihood of contralateral breast cancer diagnosis, adjusted for age, BMI, family history, and menopausal status. RESULTS The DL model achieved unweighted and weighted Dice scores of 0.88 (standard deviation [SD] = 0.08) and 0.76 (SD = 0.15), respectively, on the held-out test set, demonstrating good agreement between the model and 3D reference segmentations. There was a significant association between the odds of breast cancer diagnosis and model-derived VBD (odds ratio [OR], 1.41 [95 % CI, 1.13 to 1.77]; P = .002), with an AUC of 0.65 (95% CI, 0.60 to 0.69). ADV was also significantly associated with breast cancer diagnosis (OR, 1.45 [95% CI, 1.22 to 1.73]; P < .001) with an AUC of 0.67 (95% CI, 0.62 to 0.71). CONCLUSION DL-derived density measures derived from 3D reconstructed DBT images are associated with breast cancer diagnosis.

Abstract Glioblastoma, the most common malignant primary adult brain tumor, poses significant diagnostic and treatment challenges due to its heterogeneous molecular and micro-environmental profiles. To this end, we organize the BraTS-Path challenge to provide a public benchmarking environment and a comprehensive dataset to develop and validate AI models for identifying distinct histopathologic glioblastoma sub-regions in H&E-stained digitized tissue sections. We identified 188 multi-institutional diagnostic slides of glioblastoma (IDH-wt, Gr.4) cases, from the TCGA-GBM and TCGA-LGG data collections, following their reclassification according to the 2021 WHO classification criteria. Sub-regions were selected according to distinctive morphology of histopathologic features and included aggressive tumor biology and areas consistent with potential treatment effect. Selected sub-region annotations included cellular tumor, geographic necrosis, cortical infiltration, pseudopalisading necrosis, microvascular proliferation, white matter penetration, regions dense with macrophages, leptomeningeal infiltration, and presence of lymphocytes. We obtained 107,340 patches of size 512x512 from the 9 sub-regions. A global network of board-certified expert neuropathologists defined and followed a systematic annotation protocol based on clinical definitions and only delineated sub-regions with high confidence, thus ensuring high-quality standardized data. Each tissue section was assigned to an annotator-approver pair, with the annotator delineating sub-regions and the approver ensuring the consistency of the annotations. By crowdsourcing annotations, the BraTS-Path challenge harnesses the collective expertise of clinical neuropathologists and fosters a collaborative environment to advance the neuro-oncology field. The anticipated developed algorithms are expected to integrate state-of-the-art computational methods, achieving high accuracy in identifying diverse histopathologic features and advancing clinical decision-making processes. The BraTS-Path challenge aims to bridge the gap between research and clinical practice by promoting the development of AI-driven tools for precise tumor characterization. This collaborative effort can significantly enhance our understanding of glioblastoma, improve diagnostic accuracy, and inform treatment strategies, thereby contributing to better patient outcomes.

Abstract Background: It has been widely established that breast density is an independent breast cancer risk factor. With the increasing utilization of digital breast tomosynthesis (DBT) in breast cancer screening, there is an opportunity to estimate volumetric breast density (VBD) routinely. However, current available methods extrapolate VBD from 2D images acquired with DBT and/or depend on the existence of raw DBT data which is rarely archived by clinical centers due to cost and storage constraints. This study aims to harness deep learning to develop a computational tool for VBD assessment based solely on 3D reconstructed, “for presentation” DBT images. Methods: We retrospectively analyzed 1,080 negative DBT screening exams (09/20/2011 - 11/25/2016) from the Hospital of the University of Pennsylvania (mean age ± SD, 57 ± 11 years; mean BMI ± SD, 28.7 ± 7.1 kg/m2; racial makeup, 41.2% White, 54.2% Black, 4.6% Other), for which both 3D reconstructed and 2D raw DBT images (Selenia Dimensions, Hologic Inc) were available. All available standard views (left and right mediolateral-oblique and cranio-caudal views) were included for each exam, leading to 7,850 DBT views. Corresponding 3D reference-standard tissue segmentations were generated from a previously validated software that uses both 3D reconstructed slices and raw 2D DBT data to provide VBD metrics, shown to be strongly correlated with VBD measures from MRI image volumes. We based our deep learning algorithm on the U-Net architecture within the open-source Generally Nuanced Deep Learning Framework (GaNDLF) and created a 3-label image segmentation task (background, dense tissue, and fatty tissue). Our dataset was randomly split into training (70%), validation (15%) and test (15%) sets, while ensuring that all views of the same DBT exam were assigned to the same set. The performance of our deep learning algorithm against the corresponding reference-standard segmentations was measured in terms of Dice scores (DSC), with 0 signifying no overlap and 1 signifying perfect overlap, overall, as well as separately for each label. Results: After training was complete, our deep learning algorithm achieved a DSC of 0.78 on the validation, as well as on the test set. Our method accurately segmented background from breast tissue (DSC = 0.94) and demonstrated moderate to high performance in segmenting dense and fatty tissue, respectively (DSC = 0.49 and 0.89). Conclusion: Our preliminary analysis suggests that deep learning shows promise in the estimation of VBD using 3D DBT reconstructed, “for presentation” images. Future work involving transfer learning based on ground truth masks by clinical radiologists could further enhance this method’s performance. In view of rapid clinical conversion to DBT screening, such a tool has the potential to enable large retrospective epidemiologic and personalized risk assessment studies of breast density with DBT. Citation Format: Vinayak S. Ahluwalia, Walter Mankowski, Sarthak Pati, Spyridon Bakas, Ari Brooks, Celine M. Vachon, Emily F. Conant, Aimilia Gastounioti, Despina Kontos. Deep-learning-enabled volumetric breast density estimation with digital breast tomosynthesis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1929.

e13538 Background: Breast density is considered a well-established breast cancer risk factor. As quasi-3D, digital breast tomosynthesis (DBT) becomes increasingly utilized for screening, there is an opportunity to routinely estimate volumetric breast density (VBD). However, current methods extrapolate VBD from 2D images acquired with DBT and/or depend on the existence of raw DBT data, which is rarely archived due to cost and storage constraints. Using a racially diverse screening cohort, this study evaluates the potential of deep learning for VBD assessment based solely on 3D reconstructed, “for presentation” DBT images. Methods: We retrospectively analyzed 1,080 negative DBT screening exams obtained between 2011 and 2016 from the Hospital of the University of Pennsylvania (racial makeup, 41.2% White, 54.2% Black, 4.6% Other; mean age ± SD, 57 ± 11 years; mean BMI ± SD, 28.7 ± 7.1 kg/m2), for which both 2D raw and 3D reconstructed DBT images (Selenia Dimensions, Hologic Inc) were available. Corresponding 3D reference-standard tissue segmentations were generated from previously validated software that uses both 3D reconstructed slices and raw 2D DBT data to provide VBD metrics, shown to be strongly correlated with VBD measures from MRI image volumes. We based our deep learning algorithm on the U-Net architecture within the open-source Generally Nuanced Deep Learning Framework (GaNDLF) and created a 3-label image segmentation task (background, dense tissue, and fatty tissue). Our dataset was randomly split into training (70%), validation (15%) and test (15%) sets. We report on the performance of our deep learning algorithm against corresponding reference-standard segmentations for a cranio-caudal (CC) view-only subset. We also stratify our results by the two main racial groups (White and Black). Our evaluation measure was the weighted Dice score (DSC), with 0 signifying no overlap and 1 signifying perfect overlap, overall and separately for each label. Results: Our deep learning algorithm achieved an overall DSC of 0.682 (STD = 0.136). It accurately segmented the three labels of background, fatty tissue, and dense tissue, with DSC scores of 0.995, 0.884, and 0.617, respectively. DSC for White and Black women were 0.688 (STD = 0.127) and 0.680 (STD = 0.146), respectively. Conclusions: Our preliminary analysis suggests that deep learning shows promise in the estimation of VBD using 3D DBT reconstructed, “for presentation” CC view images and does not demonstrate bias among racial groups. Future work involving optimization of performance in other breast views as well as transfer learning based on ground truth masks by clinical radiologists could further enhance this method. In view of rapid clinical conversion to DBT screening, such a tool has the potential to enable large retrospective epidemiological and personalized risk assessment studies of breast density with DBT.